A Fuzzy Ensemble-Based Deep learning Model for EEG-Based Emotion Recognition

• Published in: Cognitive computation Vol. 16; no. 3; pp. 1364 - 1378
• Main Authors: Dhara, Trishita, Singh, Pawan Kumar, Mahmud, Mufti
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.05.2024; Springer Nature B.V
• Subjects: Accuracy; Advances in Deep Learning for Clinical and Healthcare Applications; Arousal; Artificial Intelligence; Computation by Abstract Devices; Computational Biology/Bioinformatics; Computer Science; Datasets; Deep learning; Electroencephalography; Emotion recognition; Emotions; Fourier transforms; Gompertz curves; Machine learning; Neural networks; Physiology; Support vector machines; Wavelet transforms; Emotion recognition; Gompertz function; Fuzzy ensemble; Electroencephalogram; AMIGOS; DEAP
• ISSN: 1866-9956; 1866-9964
• DOI: 10.1007/s12559-023-10171-2

Emotion recognition from EEG signals is a major field of research in cognitive computing. The major challenges involved in the task are extracting meaningful features from the signals and building an accurate model. This paper proposes a fuzzy ensemble-based deep learning approach to classify emotions from EEG-based models. Three individual deep learning models have been trained and combined using a fuzzy rank-based approach implemented using the Gompertz function. The model has been tested on two benchmark datasets: DEAP and AMIGOS. Our model has achieved 90.84% and 91.65% accuracies on the valence and arousal dimensions, respectively, for the DEAP dataset. The model also achieved accuracy above 95% on the DEAP dataset for the subject-dependent approach. On the AMIGOS dataset, our model has achieved state-of-the-art accuracies of 98.73% and 98.39% on the valence and arousal dimensions, respectively. The model achieved accuracies of 99.38% and 98.66% for the subject-independent and subject-dependent cases, respectively. The proposed model has provided satisfactory results on both DEAP and AMIGOS datasets and in both subject-dependent and subject-independent setups. Hence, we can conclude that this is a robust model for emotion recognition from EEG signals.